|Publication number||US7370251 B2|
|Application number||US 10/690,594|
|Publication date||May 6, 2008|
|Filing date||Oct 23, 2003|
|Priority date||Dec 18, 2002|
|Also published as||CA2414632A1, US20050047229, WO2004055829A1|
|Publication number||10690594, 690594, US 7370251 B2, US 7370251B2, US-B2-7370251, US7370251 B2, US7370251B2|
|Inventors||Benoit Nadeau-Dostie, Jean-François Côté|
|Original Assignee||Logicvision, Inc.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (10), Non-Patent Citations (3), Referenced by (38), Classifications (5), Legal Events (5)|
|External Links: USPTO, USPTO Assignment, Espacenet|
1. Field of the Invention
The present invention relates, in general, to testing of integrated circuits, and, more specifically, to a method of collecting in real time memory failure information for memories tested using an embedded memory test controller.
2. Description of Related Art
Memory failure information can be used to monitor and improve the quality of an integrated circuit manufacturing process. This is done by diagnosing and correlating functional failures to defects in the circuit introduced during manufacturing. The defects can in turn be associated with a certain step of the manufacturing process. Modifications can be made to this manufacturing step by using different settings for temperature, duration, dust control, and the like. If these modifications are not sufficient or possible, the circuit design might have to be changed to be more tolerant of this type of defect. One type of design modification is the introduction of redundant memory cells that can be substituted for defective memory cells.
Conventional diagnosis methods transfer detailed information about all failures off-chip to a tester for analysis. It is difficult or even impossible to do so in real time because of the bandwidth required. There are two aspects of the bandwidth problem.
A first aspect relates to the amount of information to be sent. One bit of information is required for every memory bit read. Memories often have a large number of bits (16 to 256) that are read simultaneously. This number of bits is multiplied by the number of words and by the number of read operations. This large amount of detailed information could be transferred off-chip by means of a large number of pins corresponding to the number of bits in a word. However, it is not desirable or even possible to do so. As is well known in the art, it is important to minimize the number of pins in order to reduce test cost.
A second aspect of the bandwidth problem relates to the rate at which the detailed information can be exported to a tester. Memories now operate at clock rates that exceed tester interface clock rates and is one of the reasons why embedded test controllers are used to perform memory testing. Embedded test controllers can determine whether a memory is good or bad. However, raw detailed failure information cannot be transferred without transformation because the information is generated faster than a tester can accept. One possible solution is to use a demultiplexer to reduce the rate of transfer to the tester. However, this requires multiplying the number of pins used to transfer the failure information by the ratio of the memory clock rate to the tester interface clock rate. This ratio can be as high as 10.
Chen et al, in a paper entitled “Enabling Embedded Memory Diagnosis via Test Response Compression”, 19th IEEE VLSI Test Symposium (VTS 2001) (see also Chen et al. PCT Patent Application WO 01/67463 A1 published on Sep. 13, 2001 for “Method and Apparatus for Diagnosing Memory using Self-Testing Circuits”), disclose a compression technique which uses a 6-bit output per group of fail vector. Bits of the fail vectors are combined in various ways along rows (AND, OR, 2OR (2 or more failures in the word)), columns (MaskedAND, MaskedOR, Repeat) and diagonals (XOR). Primary disadvantages of the method are that the method requires high-speed outputs, and the complexity of the functions requires splitting a fail vector into many groups, thereby increasing the number of pins that need to be connected to a tester.
Schanstra et al., in a paper entitled “Semiconductor Manufacturing Process Monitoring Using BIST for Embedded Memories” and published in the Proceedings of the International Test Conference, Oct. 18-23, 1998, disclose a method which uses several registers to collect failure information during the execution of a memory test. The registers are only inspected and exported at the end of the memory test. The registers include a fault counter, a column fault capture unit, and an address capture unit for isolated faults. The drawbacks of this method are that it does not capture information respecting faulty rows and the results of a column fault capture unit are corrupted in the presence of faulty rows. The method is restricted to algorithms that use column access mode only and requires too many registers because the test controller must accumulate the failure information until the end of the test instead of sending failure information as it is available, i.e., in real time.
Clearly, a different method is needed to compress failure information that needs to be transferred without sacrificing the ability of extracting relevant failure information. The level of resolution of the information can be traded off based on the application for which the information is required. For example, for yield analysis, it is sufficient to know the failure density (e.g., the number of failures in a column or row) for any density of failures, whereas, for repair analysis, it is necessary to know the location of individual failures more precisely when the density is low. The method of the present invention supports such trade-off. As will be seen, the present invention takes advantage of the memory structure and certain characteristics of conventional memory tests to generate failure summaries.
The present invention seeks to provide a method of collecting memory failure information in real time for memories tested using an embedded memory test controller for the purpose of process monitoring, yield enhancement, redundancy analysis and bitmap generation.
An objective of the method of the present invention is to provide an embedded test infrastructure for transferring compressed failure information off-chip at a tester clock rate in real time while performing a memory test at a system clock rate. The method allows a trade-off between the number of failure patterns that can be classified according to a classification, such as those shown in
According to a first broad aspect of an embodiment of the present invention there is disclosed a method for collecting memory failure information in real time while performing a test of memory embedded in a circuit, comprising, for each column or row of a memory under test, the steps of: (a) successively conducting uninterrupted testing of each memory location of said column or row according to a memory test algorithm under control of a first clock; (b) selectively generating a failure summary on-circuit while continuing to perform said uninterrupted testing of each memory location of said column or row; and (c) transferring off-circuit said failure summary from said circuit under control of a second clock concurrently with uninterrupted testing of a next column or row in sequence.
According to a second broad aspect of an embodiment of the present invention there is disclosed a method of collecting memory failure information in real time while performing a test of memory embedded in a circuit for memory test phases that use a column or a row access mode, comprising, for each memory column or row under test: testing each memory location of said column or row according to a memory test algorithm under control of a first clock; generating on-circuit a failure summary while testing said column or row, said generating a failure summary including, for each detected failure; determining whether said detected failure is a massive failure or a non-massive failure; and, if said detected failure is a non-massive failure: classifying said detected failure according to predetermined failure types; and updating a failure mask register with results of comparisons of memory outputs and expected memory outputs; incrementing a count of each detected failure type; and storing the row or column address of the first and last failures in said column or row, respectively; upon completion of testing of said column or row, selecting a failure summary data depending upon whether a column or row was tested; and transferring said failure summary from said circuit under control of said second clock concurrently with testing of the next column or row in sequence.
According to a third broad aspect of the embodiment of the present invention, there is disclosed a memory test controller for testing a memory in a circuit, comprising means for conducting testing of each memory location of a column or row of said memory according to a test algorithm under control of a first clock in uninterrupted fashion; means for generating a failure summary while testing the column or row of said memory; and means for transferring said failure summary from said circuit via a circuit output under control of a second clock while testing a next column or rows if any, of a memory under test.
According to a fourth broad aspect of an embodiment of the present invention, there is disclosed a memory test controller for testing memory in a circuit, comprising: means for testing each memory location of a column or row of a memory under test according to a test algorithm under control of a first clock; a failure summary generator for generating a failure summary while testing a column or row of said memory, including: failure type identification means responsive to a failure mask for classifying detected failures according to predetermined failure types; counter means responsive to outputs of said failure type identification means for counting failures of each said predetermined types; failure address registers for storing the row or column address of first and last detected failures in a column or row under test; and a failure mask register for storing a failure mask containing results of comparisons of memory data outputs against expected data outputs; means responsive to phase input signals and memory access mode signals for selecting failure data to insert into said failure summary a failure summary transfer register having a bit length equal to or less than the time required to test a column or row of said memory divided by the period of said second clock; and means for transferring said failure summary from said circuit via a circuit serial output under control of a second clock while testing the next column or row, if any, of a memory under test.
In a preferred embodiment of the method of the present invention, a failure summary is generated for each column during memory test phases that use a column access mode and for each row during memory test phases that use a row access mode. Memory test phases are executed at a first clock rate. Failure summaries are transferred from a memory test controller to an external tester at a second, lower, clock rate. Failure summary transfers are synchronized with the beginning of each column or row test phase and are performed concurrently with the memory test. While a test is being performed, detected failures are categorized into failure types and a count of each type of failure is maintained. Failure address registers store the row or column address of selected failures. A test mask register indicates which memory data outputs failed a comparison with an expected data value during execution of a test on a column or row. The test mask register is initialized at the beginning of each column, in column access mode, or row, in row access mode. Failure summaries include a combination of address information, failure count information and failure mask register information. Certain fields of information may be encoded to minimize the amount of information to transfer.
These and other features of the invention will become more apparent from the following description in which reference is made to the appended drawings in which:
In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be understood by those skilled in the art that the present invention may be practiced without these specific details. In other instances, well known methods, procedures, components and circuits have not been described in detail so as not to obscure aspects of the present invention.
The present invention provides method and a test controller for collecting and transferring off-chip memory failure information in real time while performing a test of memory embedded in a circuit. In general, the method comprises, for each column or row of a memory under test, testing each memory location of the column or row according to a memory test algorithm under control of a first clock, selectively generating a failure summary on-circuit while performing the testing, and transferring the failure summary from the circuit under control of a second clock concurrently with testing of the next column or row in sequence.
A failure summary is comprised of a combination of one or more of a column or row address of a failing cell, one or more failure counts, and failure mask data.
In a preferred embodiment, the step of generating a failure summary includes classifying each detected failure according to predetermined failure types and maintaining a count of the number of failures of each of the predetermined failure types. The predetermined failure types may include a massive failure type which is a predetermined number of failures of adjacent cells in a word of the memory. The predetermined failure types may also include a non-massive failure type which includes single bit failures and multi-bit failures.
The content of a failure summary may differ for a column and a row and a failure summary may also be generated and transferred off-chip at the end of a test phase. Thus, the step of generating a failure summary may include a step of selecting failure data to incorporate into a failure summary based on the memory test phase and/or the memory access mode. This may include selecting a column failure summary having a first set of fields when the memory is accessed in column access mode, selecting a row failure summary having a second set of fields when the memory is accessed in row access mode, and a phase failure summary having a third set of fields upon completion of a test phase. In some embodiments, a failure summary is only generated during column access mode, and no failure summaries are generated during row access mode.
In the preferred embodiment of the present invention, failure summaries are serially transferred off-chip in order to minimize the number of pins required. However, the present invention contemplates transferring failure summary data fields in parallel.
The method will be better understood with reference to the drawings.
At the beginning of each test phase (step 20), optional phase summary registers (not shown) are initialized (step 22) and at the beginning testing of each column (step 24), column failure summary registers and a fail mask register contained in a comparators block, described later, are initialized or reset (step 26). A reset will typically occur at the beginning of each test phase, but could also occur several times during a test phase if failure summaries are generated for portions of a column (or row). In addition, if a column failure summary has already been generated for a previous column, the transfer of that failure summary will begin at substantially the same time that the testing of the new column begins. The fail mask register comprises a memory element corresponding to each comparator and the content of a memory element indicates whether a failure occurred at that comparator since the register was last reset.
Next, read and write operations are performed (step 28) according to a memory test algorithm under control of a first clock. The details of the test algorithm are not important. During read operations, the memory data output is compared (step 30) against an expected value. If the data is the same, the next memory location in the column is accessed (step 32) and the read, write and compare operations are repeated. If the data is different, a failure has been detected and is processed.
When a failure is detected, it is classified (step 34) according to predetermined failure types. In a preferred embodiment of the method, a determination is made as to whether the failure is a massive failure (step 36), such as, whether a predetermined number of adjacent data output bits of a word have failed. There is a number of ways to define a massive failure. This is discussed in more detail later. Massive failures are processed (step 38) differently from non-massive failures (step 40). The primary difference is that the failure mask register is updated for non-massive failures (step 42) but not updated for massive failures because a massive failure is most probably results from a defect in the memory access mechanism to the entire location and not from a defect of individual bits. The fail mask register information is updated only when a single or a few bits of a word are failing.
After processing of a failure, massive or non-massive, failure count(s) are incremented (step 44). Failure counts may include one or more of the total number of failed locations, the total number of failed locations with massive failures, and the total number of failed locations with non-massive failures. Within the last category, there may also be a separate counts for single-bit and multi-bit failures. Multi-bit failures are two or more failures in a word of the memory.
After incrementing failure count(s), the next memory location is accessed (step 32) unless all locations in the column have been accessed (step 46). When the test of a column has been completed, column failure summary is generated, possibly encoded (step 47), loaded into a transfer register, and scheduled to be transferred off-chip (step 48) under control of a second clock. In the preferred embodiment of the invention, the failure summary is transferred serially. However, the summary may also be transferred in parallel.
After completing a test phase, when the last column (or row) has been tested (step 50), a phase failure summary may be generated and possibly encoded (step 52) and transferred off-chip (step 54).
The processing of the failures might be sufficiently complex to require more than one clock cycle and, accordingly, it may be performed in parallel with access of the next location so that there is no interruption of the test.
Failure Summary Contents
The method of the present invention seeks to provide the maximum amount of failure information that can be transferred to a tester without having to interrupt a test performed at-speed, i.e., while testing the cells of a column or row. In several cases, a complete bitmap indicating the exact location of all failures can be obtained in a single pass. In cases where the density of failures is such that the exact location of each failure cannot be transferred to the tester, statistics about the failures and partial information about the location can be transferred instead. This information might still be sufficient in some applications such as process monitoring and yield analysis. However, if more information is required, the memory test can be repeated and focus on portions of the memory.
A failure summary is generated for each column during test phases that use the column access mode (fast row) and for each row during test phases that use the row access mode. The failure summary format or content might be different during column access mode phases and row access mode phases. This is because there could be significantly smaller number of words in a row than in a column in embedded memories which use very long words, thus leaving less time to transfer failure information. Another type of failure summary format could be provided at the end of test phases to report additional information, such as the row address of bad rows, the location of isolated failures, the exact count of various failures instead of ranges, and so forth. It is also possible to defer unloading this information until the end of the test at the cost of additional on-chip storage.
Address fields store the row (or column) address of failures. Since only a limited number of such addresses can be reported, a useful option is to report the first and/or last failure address in a column. This option is useful in identifying two isolated failures and the beginning and end of a group of failures in a column (partial/full column defects). Combined with the information in failure count fields, a density of failures can be determined. One bit can be appended to each address field to indicate that an error occurred at the next address so as to thereby identify two-cell failures. Only the row (or column) address is necessary since the other components (e.g. test phase and column (or row)) of an address are implicitly known from the time of transfer of the failure summary.
Failure Count Fields
Failure count fields may include counts for single-bit failures, multi-bit failures (i.e. two or more bit failures which are not massive failures) and massive failures (e.g. at least one block of four consecutive defective bits in a word fail, other definitions possible).
Failure counts are useful in interpreting the test mask register (GOIDs). For example, a high count of single-bit failures and more than one GOID set means that one of the GOIDs corresponds to a column failure and the others to isolated failures. On the other hand, a high count of multi-bit failures means that two or more columns are bad.
Reporting of the exact value of the count during a column access only can be deferred until the end of the column test. During a column access, exact values up to three could be reported and thereafter ranges, such as >25%, >50%, and so on, could be reported. This requires encoding on three bits. The encoding could be done for memory segments (maximum of 2 or 4) of the column because partial column failure may be of interest. It will be appreciated by those skilled in the art that other forms of encoding are possible.
The counts are likely to represent counts of failed locations as opposed to mis-compares. It is not uncommon to encounter test phases where more than one compare occurs per location.
In order to distinguish single-bit failures from multi-bit failures, the present invention provides a circuit such as that shown in
The massive failure type can be identified in a number of ways. The goal is to distinguish between failures which result from failures of row or column access mechanisms and failures which result from individual bits. The criterion needs to take into account partial row or partial column failures. Massive failures tend to affect consecutive bits of a word. For example, a group of four adjacent failing bits in the failure mask register could be defined as a massive failure.
Failure Mask Register (GOID fields)
The failure mask register has a memory element associated with one or more comparators with each comparator being associated with one or more memory outputs. The most accurate results are obtained when there is one memory element associated with one comparator and one memory output. A memory element of the register contains an active value (logic 1) when a failure occurred on the associated data output(s) since the last time the register was initialized. Typically, this register is initialized at the beginning of a new column, in column access mode, or the beginning of new row, in row access mode, but initialization could be more often if statistics on segments (or portions) of a column (or row) are needed.
For memories with relatively short words (e.g., less than 8 or 16 bits), it is possible to transfer the entire fail mask register concurrently with the testing of a column or row. It might even be possible to transfer the failure mask more than once for the same column.
However, for memories with longer words and operating at a relatively high speed with respect to the tester, some trade-offs might be necessary. As already mentioned, one possibility is to divide the memory outputs into several groups and generate several failure summaries in parallel.
Another possibility is to employ encoding schemes in order to avoid having to transfer an entire failure mask. This approach is based on the assumption that there are relatively few failing memory data outputs in the same column or row.
For example, consider a memory having 64 outputs, i.e. 64 bits in a word. Instead of transferring 64 bits of failure mask, one encoding scheme is to transfer a 5-bit index or number that indicates the bit position of the output that failed together with one bit to indicate whether there are more than one failing bit in the failure mask. This form of encoding is useful when a memory tends to have single-bit failures.
Alternatively, if there are 64 GOIDs in the failure mask and it is assumed that all failures are in the same group of 8 bits, a report can be made in less that two bytes. A failure summary could have three bits to identify the group that has failures, eight bits to contain the actual GOIDs of that group and one bit to indicate whether more than one group failed, for a total of 12 bits instead of 64 bits.
If more than one group has failures, there are a few choices. Only the GOIDs of one of the groups could be reported (always follow the same convention or change the convention based on the test phase to pick up all failures as the algorithm evolves). Other encoding schemes are possible.
Statistics could be computed per segment on GOIDs. For example, the number of bad GOIDs in a segment could be counted. This has the advantage of resolving some ambiguities. For example, suppose there are errors in the first column segment that affect two GOIDs and errors in the second segment that affect two GOIDs as well, but one error is common with the first segment. At the end of the column, three GOIDs will show errors. The regions of ambiguity would be reduced considerably by knowing the number of failing GOIDs on a per segment basis. The count of GOIDs is relatively inexpensive. The number of bad GOIDs can be counted serially as the next segment is being processed and/or shifting other fields of the summary. This would involve copying the GOIDs into a shadow register. It can also be arranged to have less GOIDs than bits in the other field. If there are more GOIDs than bits for the other fields, counting would have to be done by 2 or 4 bits to have the result ready in time.
The present invention provides a memory test controller which includes means for testing each memory location of a column or row of the memory according to a test algorithm under control of a first clock, means for generating a failure summary while testing a column or row of the memory; and means for transferring the failure summary from the circuit via a circuit output under control of a second clock while testing the next column or row, if any, of a memory under test.
Memory test controller 150 is comprised of several blocks. All blocks are controlled by a first clock (Clock). A general control block 154 interacts with all other blocks as well as with an external tester (not shown) either directly or through a circuit test access port (not shown). In general, interaction with a tester is limited to initiating a memory test and, optionally, collecting failure information. Both operations require the setting of registers (one or more groups of memory elements) of the test controller to appropriate values and reading registers containing relevant information after the execution of a memory test.
General control block 154 determines the sequence of memory read and write operations that are to be performed to test the memory. The interaction of the general control block with the R/W control 156, address generator 158 and data generator 160, as well as the connections between the memory and the test controller, are well known in the art and, accordingly, are not described herein. The description herein is limited to comparators block 162 and the failure summary generator block 164, and their interaction with the other blocks.
The sequence of read and write operations is determined by memory test algorithms which are well known in the art. These algorithms are divided into phases. Several algorithms are designed so that, during a phase, all memory locations are accessed in a column access mode or a row access mode. In column access mode, the same sequence of read and write operations is applied to all memory locations of a column, one location at a time, before another column is accessed. The sequence is repeated until all locations of the column have been accessed. Similarly, in row access mode, the same sequence of read and write operations is applied to all memory locations of a row, one location at a time, before another row is accessed. The sequence is repeated until all locations of the row have been accessed.
As already indicated, it is assumed that the that at least some phases of the memory test algorithm employ a column or row access mode. During these phases, failure summary data is generated for each column (or row) on-chip and transferred off-chip for analysis. Failure summary generator 164 receives various inputs from comparators block 162 at a system clock rate, but transfers failure summary to a tester at a tester clock rate, which is usually significantly lower than the system clock rate.
A failure address block 176 receives memory address data from address generator 158, and Phase Information (Phaselnfo) and an access mode signal, AccessMode, from general control block 154 and outputs failure address data to a failure data selector 178, which also receives Access Mode and Phase Information from general control block 158, failure mask data from failure mask register 174 and failure count information from failure counters 172. Failure data selector 178 loads summary data to be output from the circuit into a transfer register 180 and may also encode data according to a predetermined encoding scheme. Alternatively, data could be encoded prior to delivery to selector 178. The data loaded into the transfer register depends on the specific failure summary combination which was designed for the circuit, the access mode and phase of a test. The transfer register operates under control of the Clock signal and a small finite state machine (FSM) 182. The transfer register has a serial input and a serial output and a clock input which receives the Clock signal. FSM 42 includes a counter 184 which counts the number of bits which have been loaded/unloaded into/from the transfer register.
It will be understood by those skilled in the art that the method can be adapted for use with more complex phases of an algorithm that access each location multiple times during the same phase. For example, an algorithm called “bit-surround” accesses, for each reference location, the reference location itself as well as all locations surrounding it. Since all locations are used as reference, each location is accessed more than once during a single phase of the algorithm. It might be preferable to only consider the failure information related to the reference cell to simplify the generation of the failure summary during this phase.
On-chip classification of some of the double failures (2-cell, 2-column or 2-rows) shown in
For memories with multiple blocks, it is preferable to generate an individual failure summary for individual blocks. However, multiple serial outputs can be used to transfer failure summaries corresponding to different blocks at the expense of extra registers in the failure summary block.
Timing of Failure Summary Transfer
Once the start bit is detected, the failure summary data is copied to the transfer register and shifted out under control of pulses of a Shift/Hold signal generated by FSM 182 (
In the simple example of
A minimum of one serial input and one serial output is needed for each transfer register. If several controllers are used in parallel, output pins dedicated to each controller are needed. More than one output could be used for each controller to facilitate obtaining failure information respecting several memories in parallel or of multiple segments of the same memory. For example, if a memory has 32-bit words and is built as two blocks, one block containing the first 16 bits of every word and the other block containing the last 16 bits of every word, failure summaries can be generated and transmitted on two outputs, one for each block. Using multiple outputs for the same memory will maximize the probability of being able to generate a complete bitmap in a single pass at the expense of more silicon area. On the input side, the number of pins depends on whether the controllers are operated asynchronously. Asynchronous operation involves a dedicated serial input for each controller. Synchronous operation involves a pause at the end of a column (or row).
Other general statistical information could also be generated and stored in appropriate registers (not shown) and scanned out at the end of each test phase and/or at the end of the test. General statistics could include, for example, the total number of mis-compares and/or locations with mis-compares. Other statistics are possible. These general statistics could be output using the output pin associated with transfer register 180 described above or using test controller normal setup mode functions.
As indicated, a memory element 202 and associated logic circuitry 206 is provided for each memory output, labeled MemoryOut. Logic circuitry 206 includes an EXOR gate 208 which receives a MemoryOut signal and expected data signal, labeled ExpectedData. The output of gate 208, labeled FailCurrenti, is active (logic 1) if its two inputs are different. The output is applied to one input of AND gate 210 whose other input receives a compare signal from general block 154 of the test controller. The output of AND gate 210 is applied to one input of an OR gate 212 which also receives the output of memory element 202, labeled FailCumulativei. The output of the OR gate 212 is applied to an AND gate 214 which also receives a reset signal, which operates to initialize the contents of memory element 202. The FailCumulativei signal indicates whether one or more errors were detected at memory output i.
The FailCurrent signal associated with all or a sub-group of k memory outputs are applied to an OR gate 216, whose output is a signal labeled FailCurrentGlobal, which indicates whether an error was detected at one or more memory outputs or in a sub-group of memory outputs.
A logic circuit 220 is associated with memory element 204. The FailCurrentGlobal signal, output by OR gate 216, is applied to one input of an AND gate 222 whose other input is a compare signal. The output of AND gate 222 is applied to an input of OR gate 224 whose other input is the feedback output of memory element 204. The output of the OR gate is applied to one input of AND gate 226 which also receives the reset signal, labeled ResetCom. The output of AND gate 226 is applied to the input of memory element 204, whose output is a signal called FailCumulativeGlobal and indicates whether one or more failures have been detected in a group of outputs of the memory.
The outputs of each of AND gate 232 of circuit 230 are combined and applied to one of the inputs of OR gate 234. The output of OR gate 234 is a FailSingleBit signal. The inverted value of this signal is applied to one input of AND gate 236 whose other input receives the FailCurrentGlobal signal corresponding to an associated number, k, of memory outputs. The output of AND gate 236 is a FailMultiBit signal. This signal is active only of FailSingleBit is inactive and FailCurrentGlobal is active.
Inverted values of all FailCurrentGlobal signals are applied to inputs of OR gate 250 which outputs a second level of the FailCurrentGlobal signal. This signal is applied to one input of AND gate 248 which receives the inverted second sage FailSingleBit signal to produce a second stage FailMultiBit signal. The outputs of the second stage detector circuit 184 are applied to respective counters in counters block 172.
It will be understood that the circuitry shown in
The figure illustrates the results of a test which included making three passes of a test algorithm that has two phases in which the first phase uses a column access mode and the second phase uses a row access mode. It will be noted that a second or more passes are performed only if there are ambiguities due to more than one bit having failures and that one or more of the bits have a large number of failures at the same column address. The failure summary combination of
As mentioned, the first phase uses a column access mode and results in the failure summaries shown in
The second phase of the algorithm used a row access mode. During this phase, the row summaries simply consist of two parts: the number of non-massive failures and the number of massive failures, as shown in
In summary, it will be seen that the present invention provides a method and circuit which overcomes the disadvantages of the known prior art. The method generates memory failure information on-chip and transfers the information off-chip at a tester interface clock rate in real time while performing an at-speed memory tests and provides sufficient information to allow failure bitmaps to be generated. The method avoids the bandwidth problems discussed earlier and requires only a single serial output, avoiding the need to increase in the number of pins or the need for high-speed pins. Further, the method can be used with algorithms that employ column and row access modes and transfers information at the end of testing each column or row, rather than storing the failure information off-chip until the end of a test. The circuit for implementing the method is relatively simple.
Although the present invention has been described in detail with regard to preferred embodiments and drawings of the invention, it will be apparent to those skilled in the art that various adaptions, modifications and alterations may be accomplished without departing from the spirit and scope of the present invention. Accordingly, it is to be understood that the accompanying drawings as set forth hereinabove are not intended to limit the breadth of the present invention, which should be inferred only from the following claims and their appropriately construed legal equivalents.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5912901 *||Mar 24, 1997||Jun 15, 1999||International Business Machines Corporation||Method and built-in self-test apparatus for testing an integrated circuit which capture failure information for a selected failure|
|US5961653 *||Feb 19, 1997||Oct 5, 1999||International Business Machines Corporation||Processor based BIST for an embedded memory|
|US6269455||Nov 7, 1997||Jul 31, 2001||Acuid Corporation Limited||System and a method for processing information about locations of defective memory cells and a memory test with defect compression means|
|US6286115 *||Jun 29, 1998||Sep 4, 2001||Micron Technology, Inc.||On-chip testing circuit and method for integrated circuits|
|US6324657||Jun 11, 1998||Nov 27, 2001||Micron Technology, Inc.||On-clip testing circuit and method for improving testing of integrated circuits|
|US6421794 *||Mar 9, 2000||Jul 16, 2002||John T. Chen||Method and apparatus for diagnosing memory using self-testing circuits|
|US6496947 *||Oct 25, 1999||Dec 17, 2002||Lsi Logic Corporation||Built-in self repair circuit with pause for data retention coverage|
|US6550023||Oct 19, 1998||Apr 15, 2003||Hewlett Packard Development Company, L.P.||On-the-fly memory testing and automatic generation of bitmaps|
|US6643807 *||Aug 1, 2000||Nov 4, 2003||International Business Machines Corporation||Array-built-in-self-test (ABIST) for efficient, fast, bitmapping of large embedded arrays in manufacturing test|
|WO2001067463A1||Mar 9, 2001||Sep 13, 2001||Mentor Graphics Corporation||A method and apparatus for diagnosing memory using self-testing circuits|
|1||Chen et al., "Enabling Embedded Memory Diagnosis via Test Response Compression", 19th IEEE VLSI Test Symposium (VTS 2001).|
|2||Chen et al., "Test Response Compression and Bitmap Encoding for Embedded Memories in Manufacturing Process Monitoring", 2001 ITC Proceedings. p. 258.|
|3||Schanstra et al., Semiconductor Manufacturing Process Monitoring using Built-In Self-Test for embedded Memories, 1998 ITC Proceedings.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7607060 *||Sep 12, 2006||Oct 20, 2009||International Business Machines Corporation||System and method for performing high speed memory diagnostics via built-in-self-test|
|US7827445 *||Dec 19, 2007||Nov 2, 2010||International Business Machines Corporation||Fault injection in dynamic random access memory modules for performing built-in self-tests|
|US7870454||May 23, 2008||Jan 11, 2011||International Business Machines Corporation||Structure for system for and method of performing high speed memory diagnostics via built-in-self-test|
|US7954018 *||Feb 2, 2007||May 31, 2011||Rudolph Technologies, Inc||Analysis techniques for multi-level memory|
|US8074130 *||Jan 22, 2008||Dec 6, 2011||Advantest Corporation||Test apparatus|
|US8358548||Oct 19, 2009||Jan 22, 2013||Avago Technologies Enterprise IP (Singapore) Pte. Ltd.||Methods for efficiently repairing embedded dynamic random-access memory having marginally failing cells|
|US8522097 *||Mar 16, 2010||Aug 27, 2013||Qualcomm Incorporated||Logic built-in self-test programmable pattern bit mask|
|US8560891 *||Oct 18, 2012||Oct 15, 2013||International Business Machines Corporation||EDRAM macro disablement in cache memory|
|US8719807 *||Dec 28, 2006||May 6, 2014||Intel Corporation||Handling precompiled binaries in a hardware accelerated software transactional memory system|
|US9003251 *||Apr 27, 2010||Apr 7, 2015||Stmicroelectronics International N.V.||Diagnosis flow for read-only memories|
|US9021325 *||Feb 7, 2013||Apr 28, 2015||Lsi Corporation||Classifying bit errors in transmitted run length limited data|
|US9046571 *||Jul 23, 2013||Jun 2, 2015||Texas Instruments Incorporated||Tap with test compression architecture and start bit detector circuit|
|US9267991 *||Mar 31, 2015||Feb 23, 2016||Avago Technologies General Ip (Singapore) Pte. Ltd.||Classifying bit errors in transmitted run length limited data|
|US9304769||May 6, 2014||Apr 5, 2016||Intel Corporation||Handling precompiled binaries in a hardware accelerated software transactional memory system|
|US9384856||Dec 11, 2013||Jul 5, 2016||Freescale Semiconductor, Inc.||Memories having a built-in self-test (BIST) feature|
|US9514844 *||Aug 26, 2014||Dec 6, 2016||Globalfoundries Inc.||Fast auto shift of failing memory diagnostics data using pattern detection|
|US9529043 *||Apr 30, 2015||Dec 27, 2016||Texas Instruments Incorporated||Tap counting circuit scan clock prime and enable prime leads|
|US9535787 *||Feb 12, 2015||Jan 3, 2017||International Business Machines Corporation||Dynamic cache row fail accumulation due to catastrophic failure|
|US9548137 *||Jun 30, 2014||Jan 17, 2017||Intel Corporation||Integrated circuit defect detection and repair|
|US9564245 *||Dec 26, 2013||Feb 7, 2017||Intel Corporation||Integrated circuit defect detection and repair|
|US20070237527 *||Mar 31, 2006||Oct 11, 2007||Sanjay Dabral||Optical debug mechanism|
|US20080082883 *||Sep 12, 2006||Apr 3, 2008||Kevin W Gorman||System for and method of performing high speed memory diagnostics via built-in-self-test|
|US20080162886 *||Dec 28, 2006||Jul 3, 2008||Bratin Saha||Handling precompiled binaries in a hardware accelerated software transactional memory system|
|US20080189582 *||Feb 2, 2007||Aug 7, 2008||Yield Dynamics, Inc.||Analysis techniques for multi-level memory|
|US20080201621 *||Jan 22, 2008||Aug 21, 2008||Advantest Corporation||Test apparatus|
|US20080222464 *||May 23, 2008||Sep 11, 2008||International Business Machines Corporation||Structure for System for and Method of Performing High Speed Memory Diagnostics Via Built-In-Self-Test|
|US20090164846 *||Dec 19, 2007||Jun 25, 2009||International Business Machines Corporation||Fault Injection In Dynamic Random Access Memory Modules For Performing Built-In Self-Tests|
|US20110029827 *||Jul 29, 2009||Feb 3, 2011||International Business Machines Corporation||Method, apparatus, and design structure for built-in self-test|
|US20110090751 *||Oct 19, 2009||Apr 21, 2011||Avago Technologies Enterprise IP (Singapore) Pte. Ltd.||Systems and methods for efficiently repairing dynamic random-access memory having marginally failing cells|
|US20110231716 *||Apr 27, 2010||Sep 22, 2011||St Microelectronics Pvt Ltd.||Diagnosis flow for read-only memories|
|US20110231719 *||Mar 16, 2010||Sep 22, 2011||Qualcomm Incorporated||Logic Built-In Self-Test Programmable Pattern Bit Mask|
|US20130042144 *||Oct 18, 2012||Feb 14, 2013||International Business Machines Corporation||Edram macro disablement in cache memory|
|US20130311842 *||Jul 23, 2013||Nov 21, 2013||Texas Instruments Incorporated||Test compression in a jtag daisy-chain environment|
|US20140223270 *||Feb 7, 2013||Aug 7, 2014||Lsi Corporation||Classifying bit errors in transmitted run length limited data|
|US20150187436 *||Jun 30, 2014||Jul 2, 2015||Intel Corporation||Integrated circuit defect detection and repair|
|US20150187439 *||Dec 26, 2013||Jul 2, 2015||Bruce Querbach||Integrated circuit defect detection and repair|
|US20150205653 *||Mar 31, 2015||Jul 23, 2015||Avago Technologies Limited||Classifying bit errors in transmitted run length limited data|
|US20150234009 *||Apr 30, 2015||Aug 20, 2015||Texas Instruments Incorporated||Test compression in a jtag daisy-chain environment|
|International Classification||G11C29/00, G11C29/40|
|May 4, 2009||AS||Assignment|
Owner name: COMERICA BANK, MICHIGAN
Free format text: SECURITY AGREEMENT;ASSIGNOR:LOGICVISION, INC.;REEL/FRAME:022629/0938
Effective date: 20090424
|Sep 15, 2009||AS||Assignment|
Owner name: LOGICVISION, INC., CALIFORNIA
Free format text: RELEASE BY SECURED PARTY;ASSIGNOR:COMERICA BANK;REEL/FRAME:023234/0037
Effective date: 20090911
|Apr 14, 2010||AS||Assignment|
Owner name: MENTOR GRAPHICS CORPORATION,OREGON
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:LOGICVISION, INC.;REEL/FRAME:024233/0056
Effective date: 20090910
Owner name: MENTOR GRAPHICS CORPORATION, OREGON
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:LOGICVISION, INC.;REEL/FRAME:024233/0056
Effective date: 20090910
|Sep 23, 2011||FPAY||Fee payment|
Year of fee payment: 4
|Oct 27, 2015||FPAY||Fee payment|
Year of fee payment: 8